Is SMH the Semiconductor Play You Need for 2026? Here's What the Numbers Say

The AI infrastructure boom isn’t slowing down — it’s accelerating. And if you’re wondering where the real money flows in this cycle, follow the chips. The VanEck Semiconductor ETF (SMH) has become a focal point for investors seeking direct exposure to the entire semiconductor ecosystem powering artificial intelligence.

The Chip Shortage That Never Was — It Was Supply Catching Up

Here’s the thing: hyperscalers like Meta, Google, and others aren’t just building data centers for fun. According to Goldman Sachs, global AI-related data center capital expenditures are projected to hit $527 billion in 2026. That’s not a typo. Those billions flow directly to semiconductor companies — chip designers, manufacturers, foundries, equipment makers, and memory suppliers. And guess which ETF concentrates these exact players? SMH does, holding titans like Nvidia, Taiwan Semiconductor Manufacturing (TSMC), Broadcom, Micron Technology, and Advanced Micro Devices as its top five positions, accounting for nearly 50% of the fund’s assets.

The SMH Portfolio: Concentration as a Feature, Not a Bug

You might wonder: isn’t all that concentration risky? Actually, no. SMH’s top 10 holdings — which also include ASML, Lam Research, KLA, Texas Instruments, and Qualcomm — represent over 73% of total assets. But this isn’t scattered bets on random semiconductor plays. These are the dominant AI infrastructure enablers. You’re not relying on one company’s success; you’re betting on the entire value chain that the AI economy depends on.

The Track Record Speaks Louder Than Hype

Last year, SMH delivered roughly 49% returns — crushing the S&P 500’s 16.4% gain. But here’s the real story: over the past decade, SMH has annualized returns around 30.9%, compared to the S&P 500’s 12.9%. This ETF has weathered semiconductor cycles before. It’s not just riding the current AI wave; it’s built a decade-long track record of outperformance.

2026 Is Different: The Inference Inflection Point

Everyone talked about training AI models. GPUs running LLM training dominated headlines. But 2026 is when the real demand curve shifts. Inference — actually running those trained models in production — is exploding. Deloitte projects inference will account for two-thirds of total AI compute demand by 2026, up from one-third just three years ago in 2023.

Here’s why this matters for SMH: training is a one-time cost, front-loaded and episodic. Inference is recurring, scalable, and grows with adoption. The demand for GPUs, memory, networking gear, and power-efficient hardware becomes structural and sustained. That’s the semiconductor tailwind SMH captures directly.

What About Valuation?

At roughly 33 times trailing-12-month earnings, SMH trades in line with large-cap tech stocks. It’s not cheap, but it’s not an outlier either. For investors who don’t want to pick individual semiconductor stocks but need pure-play AI hardware exposure, SMH offers a reasonable entry point without the single-stock risk.

The Real Question Isn’t Whether to Buy SMH — It’s Whether You Can Afford Not To

With $527 billion flowing into AI data center capex in 2026, and that spending concentrating in semiconductor leaders that SMH holds, the ETF has structurally aligned incentives. The inference boom is just getting started. The question for 2026 isn’t what will drive semiconductor returns — it’s whether you have the exposure.